harlan-agent-kit

Review one pull request adversarially, hand permitted defects to Repair, verify the remote head, and publish the Harlan Agent Kit bot status. Use for rigorous pre-merge PR review.

Improve one Agent skill from explicit Review feedback. Use only for the agent-feedback Routine.

Weekly retro of Harlan GitHub Agent sessions. Group transcripts by goal, find repeated and wasted work, and rank fixes to the controller, prompts, skills, or model. Use for agent retro, session review, agent efficiency, or self-improvement runs.

Use Harlan's browsers with the correct identity. Use before browser testing, signed-in automation, screenshots, or Chrome connection recovery. Covers prompt-free Clients and Agent Chrome, shared Harlan Chrome, dev-browser, and tab cleanup.

Inspect GitHub Actions logs for warnings and errors, triage their causes, and repair actionable findings. Use for the ci-review Routine and its Issue work.

Refresh a collection of articles with verified sources, reviewed briefs, human writing, and crisp screenshots. Use for an editorial audit, article rewrite, or coordinated content refresh.

Create or audit COPY.md and write user-visible strings against it. Use before writing marketing copy, UI labels, error messages, empty states, meta tags, or email, and when copy has drifted from the canonical strings.

Run a site's nuxt-checkin module, interpret its Check Results and prompt items, and report production changes and actions. Use for daily check-ins, morning reports, and overnight production status.

Update repository dependencies weekly, including majors, in one pull request. Use for the dependency-updates Routine and its Issue work.

Review inbox email with Himalaya and propose actions. Use for inbox checks, inbox zero, follow-ups, awaiting replies, or chase-ups.

Choose Harlan's email address, Himalaya account, and Chrome profile. Use before reading email, drafting messages, configuring mail, or using signed-in browser sessions.

Create or audit GLOSSARY.md. Use before naming product concepts, writing user-visible terms, renaming concepts, or checking vocabulary drift and banned terms.

Manage or diagnose Harlan's local GitHub maintenance service. Use for repository monitoring, automated issue or PR work, agent activity, conflicts, and its dashboard.

Operate the Hogwild home server: which SSH host to use, which account holds sudo, what runs there, where each repo, service, route, and config lives, and how to install tools for the Agent. Use for any task that touches Hogwild over SSH.

Remove AI writing tells from publishable prose. Use for humanizing blogs, docs, release notes, tweets, email, or copy that sounds generated.

Find related gaps, opportunities, bugs, and improvements around current work. Use for what's next, anything else worth doing, related improvements, or zooming out after a task. Rank evidence-backed next steps by impact, effort, and confidence.

Review architecture for a TypeScript package without nuxt.config.ts. Use for package-native refactors, exports, workspaces, module boundaries, and testability.

Prioritize open issues by impact and difficulty. Use for backlog review, quick wins, or deciding what to work on next.

Build or polish Nuxt UI v4+ pages and design systems. Use for landing pages, dashboards, tokens, motion, or generic-looking UX.

Adversarially review a Nuxt frontend. Run it and verify its contract, UX, and visual behavior. Use for frontend review or testing.

Review architecture for a project with nuxt.config.ts. Use for Nuxt-native refactors across modules, layers, plugins, composables, server routes, Nitro, or runtime config.

Read a repository's stored performance Measurements, confirm which Benchmarks carry a Regression, and repair the cause. Use for the perf-review Routine and its Issue work.

Conform or scaffold TypeScript packages and Nuxt modules. Use for workspace catalogs, package config, CI, ESLint, Vitest, playgrounds, fixtures, or Nuxt module setup.

Challenge product scope and strategy before implementation. Use for feature planning, architecture choices, premise review, ambition, and silent-failure analysis.

Create or update a pull request from current work. Use when work should be branched, committed, pushed, submitted, shipped, or landed on an owned repository, and whenever a pull request title or description is written or revised, even if nothing is pushed.

Draw a code change or part of a codebase as an animated architecture or data-flow diagram. Use when opening or updating a pull request that touches more than one module, when asked to diagram, visualise or explain a change or a system, or when a reviewer needs the blast radius before the diff.

Triage all open pull requests in Harlan-owned repositories. Use to repair, rank, sign off, or decide merge order across the PR backlog.

Write release notes, changelogs, announcements, and upgrade guides for major or minor releases.

Use Ripast for AST-aware renames, moves, extracts, usages, imports, and CSS class migrations across TS, JS, and Vue. Trigger for mechanical changes spanning files.

Triage and repair all open Sentry issues across Harlan's sites. Use for Sentry check-ins, production error backlogs, and verified repair PRs.

Read a Site's ranked NuxtSEO actions through the nuxtseo CLI, confirm which ones this repository can fix, and repair the cause. Use for the seo-review Routine and its Issue work.

Own current work through delivery and smoke verification, finish loose ends, reconcile records, and clean task-owned Git state. Use to resume delivery, follow up after merge or deploy, check what remains, close off, wrap up, or finish up.

Apply Harlan's Effect-inspired TypeScript design principles. Use for non-trivial state, errors, dependencies, module boundaries, or API design, and after a production error, where a guard at the failure site would leave the category open.

Find tweet ideas, plan launch posts, and draft or polish tweets with code cards, stat cards, or screenshot wraps. Use for content strategy, what to post, when to post, or an X post.

Write or review unit tests through exported behavior. Use for new tests, bug-fix regression tests, coverage, replacing brittle implementation checks, or a test that went red because behaviour changed on purpose.

Read a Site's NuxtSEO field Core Web Vitals findings and lab Scan history, file what real users feel and what persists in the lab, and repair the cause. Use for the vitals-review Routine and its Issue work.